A cable head multi-source information fusion and quantitative evaluation method

Abstract

The application provides a cable head multi-source information fusion and quantitative evaluation method, comprising: acquiring multi-source data of the same cable head; obtaining aligned data sequences based on the multi-source data; constructing initial basic probability distribution of a preset level state based on the aligned data sequences; determining a state confidence vector through credibility correction and conflict processing on the initial basic probability distribution and the aligned data sequences; and determining the cable head health level based on the state confidence vector. The method provided by the application can effectively suppress the interference of low-quality data on the evaluation result, thereby effectively improving the robustness and readability of the cable head state evaluation under complex interference working conditions and enhancing the diagnosis reliability.

Description

Technical Field

[0001] This application relates to the field of power equipment condition monitoring and intelligent operation and maintenance technology, and in particular to a method for multi-source information fusion and quantitative evaluation of cable heads. Background Technology

[0002] Cable heads, as critical connecting components between medium and high voltage cable insulation systems and primary equipment, perform multiple functions including electric field equalization, mechanical sealing, and conductive connection. Typical failure mechanisms include: stress cone aging and surface discharge, dendrite growth caused by moisture intrusion at the interface, localized overheating due to shielding or sheath damage, temperature rise and harmonic anomalies caused by poor grounding terminal contact, and air gaps and electric field distortion caused by installation process deviations. Once these defects evolve, they can lead to increased partial discharge levels, hot spot formation, and abnormal sheath circuit current, ultimately inducing insulation breakdown and power outages. Because cable heads are often located inside cabinets, wells, tunnels, or in humid, high-salt environments, traditional manual inspections often fail to detect early signs in a timely manner.

[0003] Existing monitoring and diagnostic methods mainly include: (1) Infrared thermography: This method determines poor contact, overheating, or current imbalance by measuring the surface temperature rise and hot spot morphology at the termination point. This method is intuitive and low-cost, but it is easily affected by load fluctuations, wind cooling, solar reflection, mirror materials, and window transmittance. It is also not sensitive enough to deeply buried or obstructed parts. Single inspections are prone to randomness and it is difficult to provide stable risk trends.

[0004] (2) Partial discharge detection: Ultra-high frequency (UHF) and ultra-high frequency (UT) systems are commonly used to characterize the electromagnetic / acoustic features of insulation defects. UHF is very sensitive to electrical breakdown inside the insulation, especially air gap discharge or discharge in insulating oil; UT can detect not only discharges, but is also particularly sensitive to mechanical defects (such as loose clamps and core vibration). However, these systems are easily affected by on-site electromagnetic interference, mechanical noise, and structural coupling differences, and the sensitivity and indicativeness of each system are not one-to-one, with significant differences in the detectability of different defect types.

[0005] (3) Sheath / Grounding Current Monitoring: Shielding or grounding anomalies are identified by measuring the effective value (RMS) and harmonic components of the sheath circuit. Its advantage is that it can be online for a long time, but it is easily confused with systemic operating condition changes (load, harmonic source switching), and needs to be interpreted in conjunction with structural and environmental information.

[0006] Although existing monitoring methods such as infrared thermography, partial discharge detection, and sheath current detection are widely used in cable operation and maintenance, single-source sensing still has significant blind spots and uncertainties. Infrared detection is susceptible to environmental interference and has thermal inertia hysteresis, making it difficult to capture early non-thermal defects inside the insulation; while partial discharge detection is sensitive to insulation defects, it is highly susceptible to interference from complex electromagnetic environments on site, resulting in false alarms, and a single physical quantity is insufficient to comprehensively characterize the fault type.

[0007] To overcome the limitations of single-source monitoring, multi-source information fusion technology has emerged. However, existing methods mostly rely on the traditional Dempster-Shafer (DS) evidence theory. This theory has an inherent mathematical flaw when dealing with highly conflicting evidence, namely the well-known "Zadeh paradox." Specifically, when affected by sensor aging drift or external interference, different data sources may provide highly contradictory diagnostic conclusions for the same target (e.g., infrared radiation indicates "severe overheating" with high confidence, while partial discharge indicates "normal insulation"). The normalization coefficients in traditional synthesis rules then become ineffective. The result will approach 1. Forcing a fusion at this point will lead to counterintuitive errors in the calculation results (such as judging it as an irrelevant third state) or even calculation divergence.

[0008] Faced with the stringent requirements for high equipment reliability in the construction of new power systems, the current cable operation and maintenance model is undergoing a profound transformation from "post-event emergency repair" to "proactive early warning" and "condition-based maintenance." However, existing monitoring technologies are insufficient to meet this transformation requirement in terms of algorithm robustness and evaluation precision: on the one hand, the lack of dynamic evaluation and feedback mechanisms for source data quality makes the fusion model prone to failure in sensor aging or strong interference environments; on the other hand, the coarse qualitative grading results cannot support refined decision-making such as remaining life prediction. Summary of the Invention

[0009] This application provides a method for multi-source information fusion and quantitative evaluation of cable heads. To solve the above-mentioned technical problems, this application adopts the following technical methods: Firstly, this application provides a method for multi-source information fusion and quantitative evaluation of cable heads, including: Acquire multi-source data from the same cable head; Based on the multi-source data, an aligned data sequence is obtained; Based on the aligned data sequence, an initial basic probability allocation for a preset level state is constructed; For the initial basic probability assignment and the aligned data sequence, the state confidence vector is determined through confidence correction and conflict handling; Based on the state confidence vector, the health level of the cable head is determined.

[0010] Optionally, the multi-source data includes partial discharge signals, infrared thermal image data, sheath or grounding current data, and environmental and operating condition data.

[0011] Optionally, obtaining the aligned data sequence based on the multi-source data includes: The multi-source data is time-base corrected and resampled to obtain an aligned data sequence.

[0012] Optionally, the step of constructing an initial basic probability allocation for a preset level state based on the aligned data sequence includes: Set a preset level of cable head health status and a single-source classifier corresponding to the preset level; The aligned data sequence is input into the single-source classifier, which outputs the initial basic probability assignment of the preset level state.

[0013] Optionally, determining the state confidence vector by refining the initial basic probability assignment and the aligned data sequence through confidence correction and conflict resolution includes: Based on the aligned data sequence, calculate the quality vector of each data source; Based on the quality vector, determine the source credibility weight of each data source; Using the source confidence weight, the initial basic probability allocation is modified by confidence reduction, and all the confidence lost due to poor sensor quality is transferred to the uncertain state of the whole set to obtain the modified probability allocation; The state confidence vector is determined by conflict resolution for the corrected probability assignment.

[0014] Optionally, the determination of the state confidence vector through conflict resolution in the modified probability assignment includes: Based on the modified probability allocation, the conflict coefficient is determined; Determine whether the conflict coefficient is less than a preset safety threshold; If so, the Dempster synthesis rule is used to orthogonally fuse the corrected probability assignments to generate a state confidence vector; If not, then weight suppression is performed on the abnormal data source corresponding to the corrected probability allocation, and the data is re-fused based on the corrected probability allocation until the conflict coefficient is less than the preset security threshold, and a state confidence vector is generated.

[0015] Optionally, determining the cable head health level based on the state confidence vector includes: Construct a normalized state utility vector; The state confidence vector is subjected to a Pignatic probability transformation to obtain the Pignatic probability of each preset level state. The cable head health index is determined based on the Pignistic probability and the normalized state utility vector. Obtain the optimal grading threshold vector between each preset level state; The cable head health level is determined based on the cable head health index and the optimal grading threshold vector.

[0016] Optionally, the process of determining the optimal grading threshold vector among the preset level states includes the following steps: Obtain a dataset of cable heads with historically labeled states; The receiver operating characteristic (ROC) curve was used to process the cable head dataset, and the corresponding true positive rate and false positive rate were calculated respectively. The Youden index is determined based on the true positive rate and the false positive rate. Based on the Yoden index, the optimal grading threshold vector between each preset level state is determined.

[0017] Optionally, determining the cable head health level based on the cable head health index and the optimal grading threshold vector includes: Based on the cable head health index, a confidence interval is determined; Determine whether the confidence interval crosses the boundary of the optimal classification threshold vector or whether the confidence interval is greater than a preset warning value; If any of the judgment results are true, the status "to be verified" will be output, triggering supplementary detection, manual review, or re-fusion processing. If all judgment results are negative, the cable head health level is determined based on the relationship between the cable head health index and the optimal grading threshold vector.

[0018] In a second aspect, this application also provides a computer-readable medium storing computer program code that, when executed by a processor, implements the method as described in any of the first aspects.

[0019] This application has the following beneficial effects: The method proposed in this application can effectively suppress the interference of low-quality data on the evaluation results, thereby effectively improving the robustness and readability of cable head condition evaluation under complex interference conditions and enhancing its diagnostic reliability. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a method for multi-source information fusion and quantitative evaluation of cable heads, provided in an embodiment of this application. Detailed Implementation

[0021] To facilitate understanding by those skilled in the art, the present application will be further described below in conjunction with embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present application.

[0022] To solve the above technical problems, such as Figure 1 As shown, this application proposes a method for multi-source information fusion and quantitative evaluation of cable heads, including: Step S101: Obtain multi-source data for the same cable head; The multi-source data acquired in this step generally includes partial discharge signals, infrared thermal imaging data, sheath or grounding current data, and environmental and operating condition data. These multi-source data provide comprehensive and reliable data support for subsequent fault diagnosis, condition assessment, and early warning analysis from the dimensions of electrical insulation status, surface temperature field distribution, grounding circuit integrity, external environmental interference, and real-time equipment operating status.

[0023] Step S102: Based on the multi-source data, obtain an aligned data sequence; Because of the differences in the acquisition terminals, sampling frequencies, and triggering mechanisms of various monitoring data, problems such as time sequence misalignment and sampling point mismatch are very likely to occur. Therefore, it is necessary to carry out time base correction and resampling processing on multi-source data to obtain a data sequence with time alignment and unified sampling frequency, thereby ensuring the consistency of multi-source data in time and space and guaranteeing the reliability of the data.

[0024] Step S103: Based on the aligned data sequence, construct the initial basic probability allocation of the preset level state; Set the preset level of cable head health status. ( ,and ), These correspond to {Normal, Attention, Abnormal, Danger}, and a single-source classifier corresponding to the preset level state A. By inputting the aligned data sequence into the corresponding single-source classifier (such as a support vector machine or neural network), the posterior probability of each preset level state can be obtained, and then converted into an initial basic probability assignment (BPA) function. .at this time, .

[0025] Step S104: Determine the state confidence vector by performing confidence correction and conflict handling on the initial basic probability assignment and the aligned data sequence; In this step, quality perception and source credibility assessment are performed. The system does not presuppose that all sensors are available at all times; instead, it establishes a dynamic evaluation mechanism based on multi-dimensional quality characteristics. For the time-aligned... Data source sequence Calculate its source credibility weight using the following sub-steps. : First, based on the aligned data sequence, the quality vector of each data source is calculated. The quality vector generally includes the signal-to-noise ratio, data missing rate, and sensor drift. For partial discharge or high-frequency current data, an energy ratio-based calculation method is used, and a sliding time window is set to divide the data within the window into effective signal segments. ) and background noise segment ( ). Calculate the signal-to-noise ratio (SNR) Signal-to-Noise Ratio , SNR As shown in the following formula: (1) in, It is the root mean square operator, i.e. N is the total number of data source sequences.

[0026] Note: For infrared thermal imaging data, the signal-to-noise ratio can be defined as the ratio of the average temperature rise of the target area to the standard deviation of the temperature of the background area.

[0027] To assess communication transmission quality, packet loss within a unit time window is statistically analyzed, and the data missing rate is calculated. As shown in the following formula: (2) in, This is the theoretical total number of data packets calculated based on the sampling rate and window duration. This represents the actual number of complete data packets received. The value range is [0,1].

[0028] To address sensor aging or zero-point drift, calculate the deviation between the current measurement reference and the calibration reference, and calculate the sensor drift. As shown in the following formula: (3) in, This serves as the reference value for the data within the current window (such as the DC component of the power frequency cycle or a reading at ambient temperature). This refers to the sensor's factory calibration value or its historical long-term average value. This is the dimensionless normalized deviation value.

[0029] A nonlinear mapping model is constructed to transform the above indicators into confidence levels within the (0,1) interval. The Sigmoid activation function is used, as shown in the following formula: (4) in, The positive gain coefficient of the signal-to-noise ratio (SNR) SNR The higher the value, the higher the credibility. These are penalty coefficients for the missing rate and the amount of drift, respectively (the more severe the missing or drifting, the lower the confidence level). This is a bias term used to adjust the sensitivity threshold for confidence.

[0030] Subsequently, based on the source credibility weights of each data source... To correct the initial basic assignment by reducing the confidence level, the confidence level lost due to poor sensor quality is reduced. (Low), all transferred to the uncertain state of the entire set. (That is, "the system is completely unaware of this"), thus obtaining the corrected probability assignment BPA. This approach aims to address the synthesis failure problem (Zadeh paradox) of traditional DS theory under low-quality or highly conflicting evidence. The reduction process is as follows: (5) in, For the initial basic probability assignment, let represent the th Without considering its own quality issues, a sensor (such as an infrared thermal imager) can only provide a specific assessment of a particular health state. A The level of support for (e.g., "abnormalities"). This is the source credibility weight, with a value range of (0,1]. It acts as a "discount factor": when... When the sensor quality is excellent, the corrected confidence level remains almost unchanged; when When the sensor drifts or malfunctions, it will forcefully compress the initial confidence level, bringing it close to 0. Correcting the probability allocation is the true "net confidence value" input to the fusion model, eliminating false confidence caused by sensor noise. Forced uncertain conversion value. This is the core innovation of this patent; it represents the trust value deducted due to poor sensor quality. This value does not disappear but is forcibly assigned to the uncertain state. . The corrected system uncertainty consists of two parts: the original uncertainty after it has been reduced. The uncertainty caused by poor quality This increase significantly reduces the weight of low-quality sensors in subsequent fusion, thus preventing them from erroneously interfering with others. In other words, the new uncertainty equals the original uncertainty reduced by the addition of the uncertainty forcibly assigned due to poor quality.

[0031] Then, based on this corrected probability assignment, the state confidence vector is determined through conflict resolution, as follows: With two sources of evidence and For example, calculate its conflict coefficient K, which reflects the degree of conflict between pieces of evidence, as shown in the following formula: (6) These represent the sets of hypotheses supported by the first and second data sources, respectively. This refers to a consistency determination. It means that when two data sources agree (e.g., both consider it "abnormal"), their probability product will contribute positively to the final result A.

[0032] Conflict determination. This refers to a situation where two data sources have completely contradictory opinions (e.g., one says "normal" and the other says "dangerous"), and their intersection is an empty set.

[0033] The value of K is between [0, 1]. The larger the K value, the deeper the contradiction between the sources. Since the reliability of low-quality sources has already been transferred to... ( Since the intersection with any set is not empty, it can significantly reduce the value of K and avoid the denominator 1-K from approaching 0, thus solving the Zadeh paradox. This is a normalization factor. It is used to redistribute the remaining probabilities after eliminating conflicts, ensuring that the sum of the probabilities of all states is ultimately 1.

[0034] Based on the conflict coefficient K, the system performs monitoring and adaptive feedback. After calculating the value of K, the system determines whether K is less than a preset safety threshold (e.g., 0.8). like This indicates that the revised conflict of evidence is now within a manageable range (because...). The introduction of If the intersection with any set is not empty, greatly reducing the K value, then the Dempster synthesis rule is used to orthogonally fuse the corrected probability assignment to generate a state confidence vector. The fusion process is as follows: (7) In the formula, Representing the final fusion confidence level, it indicates that after combining information from multiple sources such as infrared, partial discharge, and current, the system determines the cable head to be in a certain state. A The overall confidence level.

[0035] This leads to the state confidence vector M, which is composed of different preset level states: (8) In the formula, To ascertain the probability of being in the k-th state (e.g., normal, attentive, abnormal, dangerous); This is the uncertainty confidence level (i.e., the probability of "not knowing" or "unable to judge").

[0036] State confidence vector M It is the final belief distribution of the system regarding the current state of the cable head after fusion using Dempster's synthesis rules.

[0037] like This indicates that an exceptionally strong conflict still exists. At this point, a feedback mechanism is triggered, correcting the probability allocation corresponding to the abnormal data source by applying weight suppression, further reducing the impact of this conflict across all evidence sources. The weight of the smallest (worst quality) source is re-fused based on the corrected probability allocation until the conflict coefficient is less than the preset safety threshold or the preset iteration termination condition is met, generating a state confidence vector.

[0038] Step S105: Determine the health level of the cable head based on the state confidence vector.

[0039] This step involves implementing the quantification of the Health Index (HI) and multi-level state assessment. The aim of this step is to establish a nonlinear mapping mechanism from the multidimensional evidence space (DS theory output) to the one-dimensional decision space (health index).

[0040] To quantify the impact of different states on device lifespan, a Normalized State Utility Vector is constructed. : (9) In the formula, : represents the utility score for the i-th state, and It should satisfy the monotonically decreasing property, that is This is used to characterize the gradual decline in health levels.

[0041] To address the cable head health index (HI), a point estimation based on Pignistic probability transformation is introduced. The existence of uncertainty makes direct decision-making difficult. To address this, the Pignistic probability transformation from the Transmissible Belief Model (TBM) is introduced. This transformation applies a Pignistic probability transformation to the state confidence vector M, uniformly distributing the uncertainty across each single-point hypothesis to obtain the Pignistic probability for each preset level of state. : (10) Where N is the total number of state levels (N=4 in this application). This transformation eliminates The resulting ambiguity led to the construction of a normalized probability measure, which provided a mathematical basis for subsequent expectation calculations.

[0042] Based on this, the cable head health index HI is defined as the mathematical expectation of this probability distribution: (11) Among them, the cable head health index HI compresses the multidimensional probability distribution into a continuous scalar through weighted summation. Furthermore, it embodies Laplace's "principle of insufficient reason," which assumes that the system is equally likely to be in each state when information is lacking, thereby avoiding radical misjudgments.

[0043] Based on the above, in order to demonstrate the reliability of the quantitative fusion results, the upper and lower probability limits of the DS theory are used to construct the HI confidence interval for the health index. : (12) (13) in, This represents the worst-case scenario, assuming all uncertainties. In fact, all of these belong to the most severe fault state (utility is 0); conversely To best estimate the situation, assume all uncertainties. In fact, all of these represent the most ideal state of health (utility of 1).

[0044] Then the interval width This width is directly used as an indicator of system reliability: if If the value is too high, it indicates that the monitoring system has not only failed to identify the state, but has also introduced a lot of uncertainty. In this case, automatic decision-making should be rejected.

[0045] Furthermore, to overcome the subjectivity of traditional empirical threshold methods, an adaptive threshold tuning based on ROC (Receiver Operating Characteristic Curve) optimization is adopted for the cable head health index to improve the reliability of HI values. The optimal grading threshold vector between each preset level state is determined using the Receiver Operating Characteristic Curve (ROC) and the Youden Index. .

[0046] First, select the historically labeled cable head dataset. This forms a positive sample set P (the true state is "normal") and a negative sample set N (the true state is "abnormal"), and then iterates through the ROC space for optimization. Plot the ROC curves for the true positive rate (TPR) and the false positive rate (FPR) at all possible cutoff points within the interval. And the actual proportion of normal / abnormal samples: (14) (15) In the formula, It is a true positive. The true positive rate, It was a false positive. This represents the false positive rate.

[0047] This leads to the Yoden Index. Select The value of t at which the maximum value is reached is taken as the optimal threshold. : (16) Similarly, by adjusting the definitions of positive and negative samples, the ability to distinguish between "attention" and "abnormality" can be calculated separately. And "abnormal / dangerous" .

[0048] Before determining the health level of the cable head, it is necessary to check whether the confidence interval is too wide or crosses the boundary, that is, to determine whether the confidence interval crosses the boundary of the optimal classification threshold vector. (i=1, 2, 3) or whether the confidence interval is greater than the preset warning value. ( (Assuming a preset warning value), if any of the judgment results is "yes," the circuit breaker mechanism is triggered, outputting a status pending verification message, and initiating supplementary testing, manual review, or re-fusion processing. If all judgment results are "no," the cable head health index HI is then placed into the interval segmented by the threshold vector for positioning. Logical criteria and level mapping are then performed accordingly. Interval 1 (Normal Zone): If The output is Class I, indicating that the equipment is healthy, has good insulation performance, and has no obvious discharge or overheating.

[0049] Interval 2 (Note area): If The output is Level II, indicating early signs of aging, such as slight PD activity, but not yet at a dangerous level. The inspection cycle should be shortened.

[0050] Interval 3 (Outlier Zone): The output is Level III, indicating a significant defect, such as a significant partial discharge or abnormal temperature rise. It is recommended to schedule a power outage to eliminate the defect.

[0051] Section 4 (Danger Zone): Output Class IV, insulation is on the verge of breakdown, posing a great risk of failure, triggering emergency tripping or immediate intervention.

[0052] In summary, the method proposed in this application can more fully utilize multi-source monitoring information of cable heads, reduce the impact of single data anomalies and complex interference conditions on the assessment results, and improve the stability, robustness, and diagnostic reliability of the condition assessment. Furthermore, this application transforms qualitative judgments into vectorized assessments, making the assessment results more intuitive, continuous, and easy to compare. It also avoids outputting unreliable conclusions when uncertainty is high, thereby improving the credibility and engineering application value of cable head health status assessments.

[0053] The above embodiments are preferred implementations of this application. In addition, this application can be implemented in other ways. Any obvious substitutions without departing from the concept of this technical solution are within the protection scope of this application.

[0054] To facilitate understanding by those skilled in the art of the improvements made by this application compared to the prior art, some of the accompanying drawings and descriptions have been simplified, and for clarity, some other elements have been omitted from this application. Those skilled in the art should realize that these omitted elements may also constitute the content of this application.

Claims

1. A method for multi-source information fusion and quantitative evaluation of cable heads, characterized in that, include: Acquire multi-source data from the same cable head; Based on the multi-source data, an aligned data sequence is obtained; Based on the aligned data sequence, an initial basic probability allocation for a preset level state is constructed; For the initial basic probability assignment and the aligned data sequence, the state confidence vector is determined through confidence correction and conflict handling; Based on the state confidence vector, the health level of the cable head is determined.

2. The method according to claim 1, characterized in that, The multi-source data includes partial discharge signals, infrared thermal image data, sheath or grounding current data, and environmental and operating condition data.

3. The method according to claim 2, characterized in that, The process of obtaining an aligned data sequence based on the multi-source data includes: The multi-source data is time-base corrected and resampled to obtain an aligned data sequence.

4. The method according to claim 3, characterized in that, The step of constructing an initial basic probability allocation for a preset level state based on the aligned data sequence includes: Set a preset level of cable head health status and a single-source classifier corresponding to the preset level; The aligned data sequence is input into the single-source classifier, which outputs the initial basic probability assignment of the preset level state.

5. The method according to claim 4, characterized in that, The determination of the state confidence vector based on the initial basic probability assignment and the aligned data sequence through confidence correction and conflict handling includes: Based on the aligned data sequence, calculate the quality vector of each data source; Based on the quality vector, determine the source credibility weight of each data source; Using the source confidence weight, the initial basic probability allocation is modified by confidence reduction, and all the confidence lost due to poor sensor quality is transferred to the uncertain state of the whole set to obtain the modified probability allocation; The state confidence vector is determined by conflict resolution for the corrected probability assignment.

6. The method according to claim 5, characterized in that, The process of determining the state confidence vector through conflict resolution in the modified probability allocation includes: Based on the modified probability allocation, the conflict coefficient is determined; Determine whether the conflict coefficient is less than a preset safety threshold; If so, the Dempster synthesis rule is used to orthogonally fuse the corrected probability assignments to generate a state confidence vector; If not, weight suppression is applied to the abnormal data source corresponding to the corrected probability allocation, and the data is re-fused based on the corrected probability allocation until the conflict coefficient is less than the preset security threshold or the preset iteration termination condition is met, and a state confidence vector is generated.

7. The method according to claim 1, characterized in that, Determining the cable head health level based on the state confidence vector includes: Construct a normalized state utility vector; The state confidence vector is subjected to a Pignatic probability transformation to obtain the Pignatic probability of each preset level state. Based on the Pignistic probability and the normalized state utility vector, the cable head health index is determined; Obtain the optimal grading threshold vector between each preset level state; The cable head health level is determined based on the cable head health index and the optimal grading threshold vector.

8. The method according to claim 7, characterized in that, The process of determining the optimal grading threshold vector between the preset level states includes the following steps: Obtain a dataset of cable heads with historically labeled states; The receiver operating characteristic (ROC) curve was used to process the cable head dataset, and the corresponding true positive rate and false positive rate were calculated respectively. The Youden index is determined based on the true positive rate and the false positive rate. Based on the Yoden index, the optimal grading threshold vector between each preset level state is determined.

9. The method according to claim 7, characterized in that, The process of determining the cable head health level based on the cable head health index and the optimal grading threshold vector includes: Based on the cable head health index, a confidence interval is determined; Determine whether the confidence interval crosses the boundary of the optimal classification threshold vector or whether the confidence interval is greater than a preset warning value; If any of the judgment results are true, the status "to be verified" will be output, triggering supplementary detection, manual review, or re-fusion processing. If all judgment results are negative, the cable head health level is determined based on the relationship between the cable head health index and the optimal grading threshold vector.

10. A computer-readable medium, characterized in that, The system contains computer program code that, when executed by a processor, implements the method as described in any one of claims 1 to 9.

Images